clausal ellipsis and case (mis)matching in icelandic

Clausal Ellipsis and Case (Mis)Matching in IcelandicJim Wood,1 Matthew Barros,1 Einar Freyr Sigurðsson2

1Yale University and 2University of Pennsylvania

AbstractIn this paper, we take a detailed look at clausal ellipsis in Icelandic, a hithertounderstudied phenomenon. We focus on case-matching facts that seem to suggesttwo things. First, robust case-matching effects suggest that clausal ellipsis requiressome amount of island repair. Secondly, and perhaps even more interestingly, con-strained instances of case-mismatching strongly suggest that there must be silentstructure in the ellipsis site. After outlining these patterns in some detail, we pro-vide a brief discussion of what an analysis of clausal ellipsis in Icelandic mustlook like.

1 Introduction

In one of the earliest papers taking a generative approach to the study of Icelandic,

Thráinsson (1975) focused on gapping constructions of the sort in (1) (strike-

through represents elided material).

(1) a. SigurðurSigurður.NOM

treystirtrusts

áon

GuðmundGuðmundur.ACC

tilfor

aðto

ratafind

íto

skólannschool

ogand

SigurðurSigurður.NOM

treystirtrusts

áon

HörðHörður.ACC

tilfor

aðto

ratafind

heim.home

‘Sigurður is depending on Guðmundur to find the way to school, andSigurður is depending on Hörður to find the way home.’

b. SigurðurSigurður.NOM

treystirtrusts

áon


tilfor

aðto

ratafind

íto

skólannschool

ogand

HörðurHörður.NOM

treystirtrusts

áon


tilfor

aðto

ratafind

heim.home

‘Sigurður is depending on Guðmundur to find the way to school, andHörður is depending on Guðmundur to find the way home.’

Thráinsson’s primary focus was on the fact that gapping of the sort in (1a) is pos-

sible in both English and Icelandic, whereas gapping of the sort in (1b) is rejected

Working Papers in Scandinavian Syntax 96 (2016), 49–90.

50

by many English speakers.

Since then, however, not much research has been done on Icelandic ellip-

sis constructions, despite the explosion of work on ellipsis in recent decades. E.F.

Sigurðsson and Stefánsdóttir (2014) briefly discuss fragment answers/responses.

Norris et al. (2014) briefly discuss noun phrase ellipsis. Platzack (2008) briefly

discusses the absence of VP-ellipsis (and VP-topicalization) in Icelandic; see also

Thoms (2012). Gengel (2007) has a fairly extensive discussion of pseudogapping

in Icelandic.1 Ott (2014) and Ott and de Vries (2016) argue that contrastive left-

dislocation and right dislocation in Icelandic and related languages should be ana-

lyzed as clausal ellipsis (essentially on par with sluicing and fragment responses).

But overall, ellipsis phenomena has been very much in the background in the Ice-

landic syntax literature.2

With respect to clausal ellipsis, the subject which we will study here, it

turns out that Icelandic is of substantial general interest. On the one hand, case-

matching facts seem to suggest that clausal ellipsis requires some amount of is-

land repair, a conclusion that has been controversial in the literature. On the other

hand, constrained instances of case-mismatching strongly suggest that there must

be silent structure in the ellipsis site, another controversial conclusion. The goal of

the present paper, then, is to introduce the basic facts of Icelandic clausal ellipsis,

outline their theoretical relevance, and briefly outline what an account of Icelandic

clausal ellipsis must look like.

The paper is structured as follows. In section 2, we provide a brief back-

ground on clausal ellipsis and the relevance of case-matching to the phenomenon.

In section 3, we present a variety of basic data, showing that Icelandic clausal el-

lipsis looks basically like what we would expect from other languages. In section

4, we show that robust case-matching facts seem to point quite strongly to the con-1Thanks to Gísli Rúnar Harðarson for pointing this out.2There has been considerably more work on null arguments, which in some cases could be

considered a kind of ellipsis (cf. Rögnvaldsson 1982, 1990, 1993; H.Á. Sigurðsson and Egerland2009), but we set aside that matter here.

51

clusion that such ellipsis repairs island-violations, in consonance with claims since

Ross (1969), and contra Merchant (2001) (for a subset of island types), Fukaya

(2007), Barros (2012), Barros et al. (2013), Barros et al. (2014). In section 5, we

discuss instances of case mismatching, which are shown to be possible under cer-

tain constrained circumstances. In section 6, we outline the implications of the

Icelandic facts for the broader theory of case-marking and ellipsis. Section 7 con-

cludes.

2 Background

Clausal ellipsis is where the sentential part of an utterance (i.e., IP, S, or TP depend-

ing on one’s preferred terminology) goes missing from the speech signal, leaving

some sub-part of the sentence overt. In (2), we have a simple case of sluicing in En-

glish, where a Wh-question goes missing from the speech signal, leaving just the

Wh-phrase overt (called the ‘remnant,’ adopting Merchant’s 2001 terminology).

The remnant undergoes Wh-movement as usual to the left periphery followed by

TP ellipsis.

(2) Jack saw someone, but I don’t know [CP whoi [TP Jack saw ti ]].

Following Merchant (2004) and Griffiths and Lipták (2012), fragments receive the

same analysis, with the pronounced material undergoing A′-movement to the left

periphery prior to TP deletion:3

(3) A: Who did Jack see?B: [CP Sallyi [TP Jack saw ti]].

Here, we survey the empirical landscape in Icelandic, reproducing the various sub-

types of sluicing and fragments which have been attested in other languages with

clausal ellipsis. We discuss the implications of the Icelandic facts for extant the-3See Hankamer (1971), Morgan (1973), for non-movement precedents where the fragment is

pronounced in-situ, with the rest of the clause undergoing non-constituent deletion.

52

ories of clausal ellipsis, paying extra attention to what have been called ‘case-

matching’ effects.

In clausal ellipsis, the remnant typically corresponds, in some intuitive sense,

to a (typically) indefinite phrase in the antecedent, called the ‘correlate.’ In (2), the

correlate for who is someone and, in (3), the correlate for Sally is who. Ross (1969)

was the first to note that in sluicing, the remnant and correlate must match in case.

We will refer to this as the ‘Case-Matching Generalization’ (CMG). The CMG is

detectible in languages that overtly mark case on nominals, illustrated below with

a German sluice. Merchant 2004 shows the same facts hold for fragments. German

schmeicheln ‘flatter’ assigns dative case to the correlate, whereas loben ‘praise’

assigns accusative; in (4a)–(4b) we see that the remnant must bear whichever case

its correlate does.

(4) a. Erhe

willwants

jemandemsomeone.DAT

schmeicheln,flatter

aberbut

siethey

wissenknow

nicht,not

{*wer{*who.NOM

/*wen/*who.ACC

/wem}./who.DAT}

‘He wants to flatter someone, but they don’t know who.’b. Er

hewillwants

jemandensomeone.ACC

loben,praise

aberbut

siethey

wissenknow

nicht,not

{*wer{*who.NOM

/wen/who.ACC

/*wem}./*who.DAT}

‘He wants to praise someone, but they don’t know who.’(Merchant, 2001, 89)

The CMG is standardly taken as evidence for the presence of unpronounced syn-

tactic material in ellipsis, as opposed to ‘interpretive’ approaches, which reject this

assumption (Ginzburg and Sag 2000; Culicover and Jackendoff 2005; Barker 2013;

Jacobson 2013). Under the assumption that the remnant is extracted from fully

present, though unpronounced, syntactic structure, we expect its case to match that

of the correlate, since they both share identical base positions at the relevant level

of representation (5a)–(5b). On the other hand, an interpretive theory must stipu-

53

late case matching, perhaps as an anaphoric property of remnants. (See the above

cited literature for various implementations.)

(5) a. Siethey

wissenknow

nicht,not

{*wer{*who.NOM

/*wen/*who.ACC

/wem}/who.DAT}

erhe

schmeichelnflatter

will.wants

b. Siethey

wissenknow

nicht,not

{*wer{*who.NOM

/wen/who.ACC

/*wem}/*who.DAT}

erhe

lobenpraise

will.wants

(Merchant, 2001, 90)

These assumptions make an interesting prediction in languages where case-

alternations are available in what otherwise appears to be the same syntactic posi-

tion, like Icelandic. In short, all else being equal, we might expect to see violations

of the CMG in sluicing and fragments in these languages. However, in recent work

on case mismatches in sluicing in Hungarian, Nykiel and Sag (2012) (citing Jacob-

son 2013) note that case-alternations in Hungarian fail to license case-mismatch in

sluicing.

(6) a. MariMary

segítetthelped

egya

{{

fiunakboy.DAT

//

fiutboy.ACC

}.}

‘Mary helped a boy.’b. Mari

Marysegítetthelped

egya

fiunak,boy.DAT

debut

nemnot

tudom,I.know.DEF

hogyQ

{{

kinekwho.DAT

//

*kit*who.ACC

}.}

‘Mary helped a boy, but I don’t know who.’

As discussed in Jacobson (2013), for at least some speakers, the alternants differ

slightly in meaning, which might mean that this paradigm resembles the one found

with Icelandic direct object case mismatches, discussed further in section 5 below.

We will show there that such mismatches are generally degraded, at least for many

speakers. However, we will also illustrate in section 5 that Icelandic does tolerate

case mismatches under clausal ellipsis in some cases. We argue that such data

54

automatically follow from “silent structure” approaches to elliptical phenomena,

in further support of the standard assumptions about case matching effects.4

3 Icelandic Sluicing and Fragment Responses: Basic Data

In this section we show that Icelandic is like other languages within which clausal

ellipsis has been attested. We show here that well known sub-types of sluicing and

fragments are found in Icelandic, which is as expected if what appears to be clausal

ellipsis in Icelandic actually is.

3.1 Basic Sluicing

In (7), we show some basic examples of sluicing in Icelandic. What we take to be

a “basic” sluice in Icelandic is a sluice with a nominal Wh-phrase remnant with

an explicit indefinite correlate argument in the antecedent.5 (7a)–(7c) show that,

generally, the remnant has to match the correlate in case. (We will return to excep-

tions to the CMG in section 5.) (7b)–(7c) show that for subjects, it does not matter

if the subject is low, as in an unaccusative expletive construction (7c), or high, in

the ordinary subject position (7b). (7d) shows (unsurprisingly) that sluicing is fine

when nominative and accusative are syncretic.6 (7e) shows that case matching is

required for dative objects as well.

4To our knowledge, case matching effects are robustly attested in languages with overt casemarking. However, some counterexamples have been uncovered thus far in a few languages. Ince(2012) notes that Turkish genitive correlates correspond to nominative remnants obligatorily insluicing; Barros (2014) and Thoms (2015) uncover abstract Case mismatches in English sluices; Vi-cente’s (2015) short survey cites counterexamples attested in Mongolian, Korean, Uzbek, Japanese,German and Chamorro (though it has been argued for some of these, namely Japanese and Uzbek,that what appears to be sluicing is actually a reduced copular clause, so that the relevance of theselanguages to the status of the CMG is questionable).

5Correspoding to the “merger” cases of Chung et al. (1995).6Syncretism in case matching has been appealed to on occasion as a licensing context for syn-

tactic mismatches between the antecedent structure and the elided clause; see especially van Crae-nenbroeck 2012 and references therein for discussion.

55

(7) a. JónJohn.NOM

sásaw

einhvern,someone.ACC,

enbut

égI.NOM

veitknow

ekkinot

{{

*hver*who.NOM

//

hvernwho.ACC

//

*hverjum*who.DAT

}.}

‘John saw someone, but I don’t know who.’b. Einhver

someone.NOM

fór,left,

enbut

égI.NOM

veitknow

ekkinot

{{

hverwho.NOM

//

*hvern*who.ACC

//

*hverjum*who.DAT

}.}

‘Someone left, but I don’t know who.’c. Það

EXPL

fórleft

einhver,someone.NOM,

enbut

égI

veitknow

ekkinot

{{

hverwho.NOM

//

*hvern*who.ACC

//

*hverjum*who.DAT

}.}

‘Someone left, but I don’t know who.’d. Jón

John.NOM

borðaðiate

eitthvað,something.NOM/ACC,

enbut

égI

veitknow

ekkinot

{{

hvaðwhat.NOM/ACC

//

*hverju*what.DAT

}.}

‘John ate something, but I don’t know what.’e. Jón

John.NOM

breyttichanged

einhverju,something.DAT,

enbut

égI.NOM

veitknow

ekkinot

{{

*hvað*what.NOM/ACC

//

hverjuwhat.DAT

}.}

‘John changed something, but I don’t know what.’

The same basic fact holds for fragment responses. We illustrate this with an

accusative direct object in (8), a dative direct object in (9), and a dative indirect

object in (10). We will discuss subjects (non-nominative subjects in particular) in

more detail in section 5.

(8) A: JónJohn.NOM

sásaw

bílinn.car.the.ACC

‘John saw the car.’

56

B: {{

*Rútan*coach.the.NOM

//

Rútunacoach.the.ACC

//

*Rútunni*coach.the.DAT

}}

líka.too

‘The coach too.’

(9) A: Höfundurinnauthor.the.NOM

breyttichanged

byrjuninni.beginning.the.DAT

‘The author changed the beginning.’B: {

{*Endirinn*ending.the.NOM

//

*Endinn*ending.the.ACC

//

Endinumending.the.DAT

}}

líka.too

‘The ending too.’

(10) A: JónJohn.NOM

gafgave

mérme.DAT

bókina.book.the.ACC

‘John gave me the book.’B: {

{*Ég*I.NOM

//

*Mig*me.ACC

//

Mérme.DAT

}}

líka.too

‘Me too.’ (I.e. ‘He gave it to me too.’)

3.2 Sprouting

Sprouting describes a situation where the remnant of a sluice lacks a correlate.

Sprouting is possible in Icelandic just as it is in other languages with sluicing.

(11) JónJohn.NOM

fór,left

enbut

égI

veitknow

ekkinot

{{

meðwith

hverjumwhom.DAT

//

hvenærwhen

//

hvertwhere.to

//

hvernighow

//

hvers vegnawhy

//

af hverjuwhy

}.}

‘John left, but I don’t know {with whom/when/where to/how/why}.’

When the sluice remnant is a DP, it must be case-marked with whatever case would

have been expected from the verb in the antecedent clause. (12a) shows this with an

ordinary inanimate object, which is case-syncretic for nominative and accusative.

(Note that the verb borða ‘eat’ takes an object in the accusative case in the active.)

(12b) shows this for an animate object, which is not case-syncretic. (If it helps, one

can imagine that Jón is a people-eating troll.)

57

(12) a. JónJohn.NOM

borðaði,ate

enbut

égI

veitknow

ekkinot

{{

hvaðwhat.NOM/ACC

//

*hverju*what.DAT

}.}

‘John ate, but I don’t know what.’b. Jón

John.NOM

borðaði,ate

enbut

égI

veitknow

ekkinot

{{

*hver*who.NOM

//

hvernwho.ACC

//

*hverjum*who.DAT

}.}

‘John ate, but I don’t know who.’

Sprouting is also possible for fragment responses, as illustrated in (13).

(13) A: JónJón

borðaðiate

loksins.finally

‘John finally ate.’B: Í

inalvöru?seriousness

‘Really?’A: Já,

Yes,{{

*ávextir*fruit.NOM

//

ávextifruit.ACC

//

*ávöxtum*fruit.DAT

}.}

‘Yes, fruit.’

3.3 SWIPING

SWIPING describes sluicing where the remnant is a prepositional phrase where

the word order of the prepositional object and the preposition are inverted from the

canonical order.7 An example from English is given in (14) below.

(14) John left, but I don’t know who with.

Ross (1969) originally analyzed this sort of word order in sluicing as non-constituent

deletion. Abstracting away from the details of Ross’s original analysis and frame-

work, this essentially gives us an analysis for the elided material in (14) as in (15):7The term is due to Merchant (2001). It stands for S(luicing) W(ith) I(nversion) of P(repositions)

in N(orth) G(ermanic).

58

(15) John left, but I don’t know who John left with.

Such an analysis suggests a correlation between the possibility of preposition strand-

ing under Wh-movement in a given language, and the availability of SWIPING un-

der ellipsis. As Merchant (2002) illustrates, SWIPING is unavailable in Icelandic,

Swedish, and Frisian, all of which are languages in which preposition stranding is

allowed under regular Wh-movement. This casts doubt on the relationship between

the availability of preposition stranding and SWIPING.8 Our own investigations

into Icelandic sluicing are consistent with Merchant’s results:

(16) a. * JónJon.NOM

fór,left

enbut

égI

veitknow

ekkinot

hverjumwho.DAT

með.with

Intended: ‘John left, but I don’t know who with.’b. * Jón

Johngerðifixed

viðP

bílinn,car.the.ACC

enbut

égI

veitknow

ekkinot

hverjuwhat.DAT

með.with

Intended: ‘John fixed the car, but I don’t know what with.’

We do not discuss this further here, except to note that our observations are con-

sistent with Merchant’s.

3.4 Contrastive Sluicing

Contrast sluices are sluices where the remnant and correlate are contrastively fo-

cused, as in (17). Unlike non-contrastive sluices, the interpretation of the correlate

and remnant must contrast in some way, shape, or form. For example, the distinc-

tion between dogs and cats is relevant in (17).

8Though it does appear to be the case that only a subset of preposition stranding languagesallow SWIPING, so perhaps the correlation is mostly correct (i.e., that perhaps there is, in fact, animplicational relationship between the availability of SWIPING, and the availability of prepositionstranding), and independent principles in languages in which this expectation is not borne out areto blame for the counterexamples mentioned here.

59

(17) Húnshe.NOM

áhas

þrjáthree

ketti,cats.ACC

enbut

égI

veitknow

ekkinot

hversuhow

margamany.ACC

hunda.dogs.ACC

‘She has three cats, but I don’t know how many dogs.’

A subtype of contrastive sluicing involves ‘else’-modification, as in the examples

in (18).

(18) a. Húnshe.NOM

kallaðicalled

HlynHlynur.ACC

fífl,fool.ACC

enbut

égI

veitknow

ekkinot

hverjawho.PL.ACC

aðra.else.PL.ACC

‘She called Hlynur a fool, but I don’t know who else.’b. Henni

her.DAT

finnstfinds

gamanfun

aðto

lesaread

íin

eldhúsinu,kitchen.the.DAT

enbut

égI

veitknow

ekkinot

hvarwhere

annarselse

staðar.place

‘She likes reading in the kitchen, but I don’t know where else.’

Contrastive sluices are interesting because they have slightly different prop-

erties than non-contrastive sluices. In particular, unlike non-contrastive sluices,

contrastive sluices are island sensitive (Fukaya 2007; Merchant 2008). See Mer-

chant (2001) especially for in-depth discussion of contrastive sluices and their con-

sequences for theories of ellipsis identity. While it would be worthwhile to see if

this holds in Icelandic as well, we must set this aside at the moment for reasons of

space and time.

3.5 Interim summary

To summarize, we find all the usual sub-types of sluicing and fragments in Ice-

landic that are found in other languages in which these constructions have been

attested. We have, furthermore, gone some way in illustrating that the known prop-

erties of these sub-types behave as expected in Icelandic. In what follows we fo-

60

cus on two construction-specific properties of clausal ellipsis as instantiated in

Icelandic, namely, the phenomenon of “island-repair” under clausal ellipsis, and

case-matching effects between remnants and correlates.

4 Potential Island Violations

It has long been observed that sluicing appears to ‘repair’ island violations (Ross

1969; Chung et al. 1995; Merchant 2001 (for some islands); Lasnik 2001; Fox and

Lasnik 2003). That is, if we understand sluicing as being derived from movement,

but deletion of what is left over, then that movement appears in some cases to vio-

late island constraints.9 In this section, we show that the same holds for Icelandic,

and that the case-matching discussed in the previous sections seems to hold in

these cases as well. We take no particular stance on the analysis of these apparent

island repair phenomena.

4.1 Relative Clauses

Sluicing appears to repair relative clause islands, as shown in (19a). (19b) shows

that relative clauses of the relevant sort are extraction islands. (19c) and (19d) show

that such cases cannot be derived by assuming that the deleted clause was really

a cleft. This addresses a vein in the literature that aims to explain the appearance

of island repair under ellipsis as illusory, stemming from non-island containing

structures in the ellipsis site, such as clefts or copular clauses (Erteschik-Shir 1977;

Fukaya 2007; Barros 2012; Barros et al. 2013, 2014). However, case-matching is

fully unacceptable for all speakers in short clefts, as shown in (19c), and most

speakers reject a long cleft as well, as shown in (19d).

9In this section, a judgment of ‘??/%’ means that speakers varied, ranging from rejecting toaccepting. Other than Einar, the judgments in this section come partially from Gísli Harðarson,Sigríður Sigurjónsdóttir, Halldór Sigurðsson and Ásgrímur Angantýsson. The sentences we askedthem about were the short and long it-cleft sentences, since the other judgments were totally clear.

61

(19) a. Þeirthey

réðuhired

einhvernsomeone.ACC

semwho

talarspeaks

sænskaSwedish

mállýsku. . .dialect.ACC

. . . en

. . . butégI

veitknow

ekkinot

hvaðawhich

{{

mállýskudialect.ACC

//

*mállýska*dialect.NOM

}.}

‘They hired someone who speaks a Swedish dialect,but I don’t know which dialect.’

b. * . . . en. . . but

égI

veitknow

ekkinot

hvaðawhich


þeirthey

réðuhired

einhvernsomeone.ACC

semwho

talarspeaks

__.__

c. . . . en. . . but

égI

veitknow

ekkinot

hvaðawhat

{*mállýsku{*dialect.ACC

/mállýska/dialect.NOM

}}

þaðit

var.was

‘. . . but I don’t know which dialect it was.’d. . . . en

. . . butégI

veitknow

ekkinot

hvaðawhat

{??/%mállýsku{??/%dialect.ACC

/mállýska/dialect.NOM

}}

þaðit

varwas

semthat

hannhe

talar.speaks

‘. . . but I don’t know which dialect it was that he speaks.’

This suggests that the apparent island-violation in (19a) cannot be explained by

assuming a cleft source. Worth noting is that these results are also consistent with

the view where the elided structure must be syntactically identical to its antecedent,

which would rule out cleft or copular sources whenever the antecedent is not itself

a cleft or copular clause.

An alternative possibility that has been explored in the literature (Merchant

2001; Fukaya 2007; Barros 2012; Barros et al. 2013, 2014) is that such cases are

derived from a “short source” such as the English example in (20a) or its Icelandic

counterpart in (20b):

62

(20) a. They hired someone who speaks a Swedish dialect, but I don’t knowwhich dialect he speaks.

b. Þeirthey

réðuhired

einhvernsomeone.ACC

semwho

talarspeaks

sænskaSwedish


enbut

égI

veitknow

ekkinot

hvaðawhich


hannhe.NOM

talar.speaks

‘They hired someone who speaks a Swedish dialect,but I don’t know which dialect he speaks.’

This possibility—at least the short source given in (20)—is, however, undermined

by sentences like (21),10 in which there is no referent that can correspond to the

pronoun in the deleted clause. That is, since no one was hired (21a), it is not pos-

sible to have something like ‘he or she speaks’ in the deleted clause.

(21) a. Þeirthey

réðuhired

ekkinot

neinnanyone

semwho

talarspeaks

ákveðnacertain

sænskaSwedish

mállýsku,dialect

enbut

égI

veitknow

ekkinot

hvaðawhich

mállýsku.dialect.ACC

‘They didn’t hire anyone who speaks a certain Swedish dialect,but I don’t know which dialect.’

b. Enginnno.one

varwas

meðwith

nemandastudent

íin

bekknumclass

sínumREFL.POSS

semwho

talarspeaks

ákveðnacertain

sænskaSwedish

mállýsku,dialect

enbut

égI

manremember

ekkinot

hvaðawhich

mállýsku.dialect.ACC‘No one had a student in their class who speaks a certain Swedishdialect, but I don’t remember which dialect.’

The same appears to hold for fragment answers, as illustrated in (22).

10Adapted from Lasnik (2001, 315) (his example (42)) in English, which makes the same point.

63

(22) A: Þeirthey

réðuhired

einhvernsomeone

semwho

talarspeaks

íslensku.Icelandic.ACC

‘They hired someone who speaks Icelandic.’B: {

{*Þýska*German.NOM

//

ÞýskuGerman.ACC

}}

líka.too

‘German too.’

This response is ambiguous. It can mean either (i) that the person they hired also

speaks German, or (ii) that they also hired someone who speaks German. The sec-

ond reading is salient in the context of a conversation where A and B are discussing

a situation where a number of people have been hired. It is reading (ii) that appears

to be an island violation, assuming that fragment responses are derived by move-

ment plus deletion. The Icelandic facts thus seem to support the view that ellipsis

can repair relative clause island violations.

4.2 Left Branch Violations

Sluicing and fragment responses also appear to violate constraints against left

branch extraction. Consider the following sentences. (23a) shows that a phrase like

hversu ríkum ‘how rich’ cannot be moved out of the phrase containing the head,

manni ‘man’. (23b), however, shows that this phrase can be stranded by sluicing.

It also shows that the case on ‘how rich’ must match the case of the associate in

the antecedent clause. Such data is troubling for the view where left branch sluices

actually stem from predicative copular clauses, with no left branch violation, as

argued for in Barros et al. (2014). In (23b), the case on the remnant must be dative.

(23c) shows that a short source like ‘. . . how rich he is’ would not be a possible

source for the sluice in (23b), since case matching is impossible in (23c). In pred-

icative sentences of this type, nominative is required.

64

(23) a. * Hversuhow

ríkumrich.DAT

giftistmarried

húnshe.NOM

manni?man.DAT

b. Húnshe.NOM

giftistmarried

ríkumrich.DAT

manni,man.DAT

enbut

égI

veitknow

ekkinot

hversuhow

{{

ríkumrich.DAT

//

*ríkur*rich.NOM

}.}

‘She married a rich man, but I don’t know how rich.’c. . . . en

. . . butégI

veitknow

ekkinot

hversuhow

{{

*ríkum*rich.DAT

//

ríkurrich.NOM

}}

hannhe

er.is

‘. . . but I don’t know how rich he is.’

Another possible source for (23b) would be to front the whole DP object, and then

elide the NP. Thus, if ellipsis of manni ‘man’ in (24a) were possible in a context

licensing sluicing, then we would not be forced to assume that there was a violation

of the left branch constraint. However, the exchange in (24b) casts doubt on this

idea, since NP-ellipsis is normally not available in a way that strands the degree

phrase and adjective phrase.

(24) a. Hversuhow

ríkumrich.DAT

manniman.DAT

giftistmarried

hún?she.NOM

‘How rich a man did she marry?’b. A: Hún

she.NOM

giftistmarried

rosalegavery

ríkumrich

manni.man

‘She married a very rich man.’B: * Hversu

howríkumrich.DAT

giftistmarried

hún?she.NOM

The data in (25) and (26) replicate the data in (23) and (24), only with an accusative

object.

65

(25) a. * Hversuhow

ríkanrich.ACC

þurftineeded

húnshe.NOM

mann?man.ACC

b. Húnshe.NOM

þurftineeded

ríkanrich.ACC

mann,man.ACC

enbut

égI

veitknow

ekkinot

hversuhow

{{

ríkanrich.ACC

//

*ríkur*rich.NOM

}.}

‘She needed a rich man, but I don’t know how rich.’c. . . . en

. . . butégI

veitknow

ekkinot

hversuhow

{{

*ríkan*rich.ACC

//

ríkurrich.NOM

}}

hannhe

er.is

‘. . . but I don’t know how rich he is.’

(26) a. Hversuhow

ríkanrich.ACC

mannman.ACC

þurftineeded

hún?she.NOM

‘How rich a man did she need?’b. A: Hún

she.NOM

þurftineeded

rosalegavery

ríkanrich.ACC

mann.man.ACC

‘She needed a very rich man.’B: ?? Hversu

howríkanrich.ACC

þurftineeded

hún?she.NOM

In (27) and (28), we show that left branches can also be stranded in fragment

responses. As before, case matching is required, and this holds for both dative and

accusative objects.

(27) A: Húnshe

giftistmarried

ríkumrich

manni.man.DAT

‘She married a rich man.’B: Já,

yesmjögvery

{{

*ríkur*rich.NOM

//

*ríkan*rich.ACC

//

ríkumrich.DAT

}.}

‘Yes, very rich.’

(28) A: Hannhe

þarfneeds

ríkarich

konu.woman.ACC

‘He needs a rich woman.’B: Já,

yesmjögvery

{{

*rík*rich.NOM

//

ríkarich.ACC

//

*ríkri*rich.DAT

}.}

‘Yes, very rich.’

66

Thus, the Icelandic facts seem to support the view that ellipsis can repair left branch

violations.

4.3 Embedded Question Island

Embedded questions are islands in Icelandic, as illustrated in (29b). The sluice in

(29a) shows that sluicing seems to fix a violation of such an island, as expected

under the island-repair view. Examples (29c)–(29d) illustrate that an account of

the apparent repair effect in terms of copular clauses fails to account for the case

matching facts. This forces us to the conclusion that we do, indeed, once again,

have apparent island repair in Icelandic sluices.

(29) a. SandraSandra.NOM

varwas

aðto

reynatry

aðto

áttafigure

sigREFL

áout

hvaðawhat

konawoman

ætlaðiintended

aðto

hittameet

ákveðinncertain

mann,man.ACC

enbut

égI

manremember

ekkinot

hvaðawhat

{{

mannman.ACC

//

*maður*man.NOM

}.}

‘Sandra was trying to figure out which woman was trying to meet acertain man, but I don’t remember which man.’

b. * . . . en. . . but

égI

manremember

ekkinot

hvaðawhat

mannman.ACC

húnshe.NOM

varwas

aðto

reynatry

aðto

áttafigure

sigREFL

áout

hvaðawhat

konawoman

ætlaðiintended

aðto

hittameet

____

c. . . . en. . . but

égI

manremember

ekkinot

hvaðawhat

{{

*mann*man.ACC

//

maðurman.NOM

}}

þaðit

var.was‘. . . but I don’t remember which man it was.’

67

d. . . . en. . . but

égI

manremember

ekkinot

hvaðawhat

{{

??/%mann??/%man.ACC

//

maðurman.NOM

}}

þaðit

varwas

semthat

konanwoman.the

ætlaðiintended

aðto

hitta.meet

‘. . . but I don’t remember which man it wasthat the woman was going to meet.’

Thus, the Icelandic facts seem to support the view that ellipsis can repair movement

out of embedded question islands.

4.4 Adjunct Island

As (30b) illustrates, adjuncts are islands to extraction in Icelandic, which appear to

be repairable under sluicing (30a). As before, (30c)–(30d) illustrate that a copular

source for the remnant fails to meet case-matching requirements (at least for those

speakers who reject such cases on long-cleft pivots).11

(30) a. JónJohn

verðurwill.be

reiðurmad

efif

SaraSara

talartalks

viðwith

einnone.ACC

afof

kennurunum,teachers.the.DAT

enbut

húnshe

geturcan

ekkinot

munaðremember

{{

hvernwhich.ACC

//

*hver*which.NOM

}}

‘John will be mad if Sara talks with one of the teachers, but shecan’t remember which.’

b. * . . . en. . . but

húnshe

geturcan

ekkinot

munaðremember

hvernwhich.ACC

hannhe

verðurwill.be

reiðurmad

efif

húnshe

talartalks

viðwith

____

c. . . . en. . . but

húnshe

geturcan

ekkinot

munaðremember

{{

*hvern*who.ACC

//

hverwho.NOM

}}

þaðit

er.is

‘. . . but she can’t remember who it is.’11Granted, it is conceivable that other sorts of copular sources aside from clefts in Icelandic yield

the right case facts under sluicing. We leave exploring this question aside here for future work.

68

d. . . . en. . . but

húnshe

geturcan

ekkinot

munaðremember

{{

??/%hvern??/%who.ACC

//

hverwho.NOM

}}

þaðit

eris

semthat

húnshe

áought

ekkinot

aðto

talatalk

við.with

‘. . . but she can’t remember who it is that she shouldn’t talk to.’

Fragments show a similar pattern. In (31), the nominative response is grammatical

only on the reading where Guðmundur will get angry—corresponding to the the

nominative correlate Jón. Under the reading where Jón gets mad if Sara talks with

Bjartur and Guðmundur, only accusative is possible on the fragment in (31B).

(31) A: JónJohn

verðurbecomes

reiðurangry

efif

SaraSara

talartalks

viðto

Bjart.Bjartur.ACC

‘John will get angry if Sara will talk to Bjartur.’B: {

{*Guðmundur*Guðmundur.NOM

//

?Guðmund?Guðmundur.ACC

//

*Guðmundi*Guðmundur.DAT

}}

líka.too

‘Guðmundur, too.’

Thus, this seems to support the view that ellipsis can repair adjunct island viola-

tions.

4.5 Coordinate Structure Constraint

Another kind of constraint that might be violable under clausal ellipsis is the Co-

ordinate Structure Constraint (CSC; Ross 1967), which says that it is impossible

to extract only one conjunct from a coordinated phrase. (32a) shows that a sluic-

ing remnant can indeed correspond to (and match the case of) one correlate in a

conjunction phrase. (32b) shows that a continuation that extracts such a conjunct

directly is ungrammatical; if such a continuation is the source for (32a), then it

appears that the illicit CSC violation is repaired by ellipsis. (32c)–(32d) show that

short and long clefts would not be possible sources for (32a), since they do not

allow case-matching.

69

(32) a. Þeirthey

sannfærðuconvinced

KennedyKennedy

ogand

einhvernsome.ACC

annanother.ACC

þingmannsenator.ACC

umon

aðto

styðjasupport

(í(in

sameiningu)unison)

frumvarpið,bill.the

enbut

égI

manremember

ekkinot

hvaðawhich

{{


//

*þingmaður*senator.NOM

}.}

‘They convinced Kennedy and some other senator to support the bill(together), but I don’t remember what senator.’

b. . . . en. . . but

égI

manremember

ekkinot

hvaðawhich


(*þeir(*they


hannhim

ogand

____

umon

aðto

styðjasupport

(í(in

sameiningu)unison)

frumvarpið)bill.the)

c. * . . . en. . . but

égI

manremember

ekkinot

hvaðawhat


þaðit

var.was

d. ??/% . . . en. . . but

égI

manremember

ekkinot

hvaðawhat


þaðit

varwas

semthat

þeirthey

sannfærðu.convinced‘. . . but I don’t remember which senator it was that they convinced.’

The same facts—i.e., case matching and apparent island repair—hold for fragment

responses. (33) replicates the context from above.

(33) A: Þeirthey


KennedyKennedy

ogand

einhvernsome.ACC

annanother.ACC


umon

aðto

styðjasupport

(í(in

sameiningu)unison)

frumvarpið.bill.the

B: Já,yes

{{

*Bjartur*Bjartur.NOM

//

BjartBjartur.ACC

//

*Bjarti*Bjartur.DAT

}.}

However, it is possible that the elided continuation does not require a CSC

violation. Depending on one’s view of the identity condition, (33B) could be de-

rived from something like (34).

70

(34) Já,yes,

BjartBjartur.ACC


þeirthey

umon

aðto

styðjasupport

frumvarpið.bill.the

‘Yes, Bjartur they convinced to support the Bill.’

This alternative is undermined by the example presented in (36). (35) shows that

a predicate like skila að ‘separate’ requires a plural internal argument. Thus, in

(36B), the fragment response would seem to have to be fed by a CSC violating

structure; it could not, for example, be derived from something like (35a), since

that is not a grammatical sentence to begin with.

(35) a. * Þeirthey

skilduseparated

BjartBjartur.ACC

að.at

b. Þeirthey

skilduseparated

BjartBjartur.ACC

ogand

PálPáll.ACC

að.at

‘They separated Bjartur and Páll.’

(36) A: Þeirthey

skilduseparated

PálPáll.ACC

ogand

einhvernsomeone

annanother.ACC

að.at

‘They separated Páll and someone else.’B: Já,

yes{{

*Bjartur*Bjartur.NOM

//

BjartBjartur.ACC

//

*Bjarti*Bjartur.DAT

}.}

‘Yes, Bjartur.’

The same goes for sluicing, as illustrated in (37).

(37) Þeirthey

skilduseparated

BjartBjartur.ACC

ogand

einhvernsome

annanother


að,at,

enbut

égI

manremember

ekkinot

hvaðawhat

þingmann.senator.ACC

‘They separated Bjartur and some other senator,but I don’t remember which senator.’

Thus, the facts presented in this section seem to support the view that ellipsis can

repair CSC violations.

71

5 Case Mismatches and Dative Substitution

Ordinarily, case mismatches are not possible in Icelandic fragment answers/responses.

This holds for verbs selecting oblique subjects, as illustrated in (38), and for verbs

that take ordinary nominative subjects, as illustrated in (39)–(40).

(38) A: {{

*Ég*I.NOM

//

*Mig*me.ACC

//

Mérme.DAT

}}

leiðist.bores

‘I’m bored.’B: {

{*Ég*I.NOM

//

*Mig*me.ACC

//

Mérme.DAT

}}

líka.too

‘Me too.’

(39) A: {{

Hverwho.NOM

//

*Hvern*who.ACC

//

*Hverjum*who.DAT

}}

skemmdibroke

sjónvarpið?television.the.ACC

‘Who broke the TV?’B: {

{ÉgI.NOM

//

*Mig*me.ACC

//

*Mér*me.DAT

}.

‘Me.’

(40) A: {{

ÉgI.NOM

//

*Mig*me.ACC

//

*Mér*me.DAT

}}

vilwant

fara.go

‘I want to go.’B: {

{ÉgI.NOM

//

*Mig*me.ACC

//

*Mér*me.DAT

}}

líka.too

‘Me too.’

Under theories of ellipsis identity where only semantic content is relevant (Mer-

chant 2001 in particular), one might have imagined that accusative or dative would

be possible in (40B), given that another verb meaning ‘want’, namely langa, takes

either an accusative subject (standardly) or a dative subject (under Dative Substi-

tution).12

12It should be noted that Dative Substitution refers to verbs that prescriptively take an accusativesubject. There is, however, a lot of both inter- and intra-speaker variation, such that some speakersmay find dative ungrammatical with these verbs, whereas others may only find dative grammatical.Many speakers, however, show intra-speaker variation. (See discussion in, e.g., Svavarsdóttir 1982;Jónsson 2003; Barðdal 2001, 2011; Jónsson and Eythórsson 2003, 2005; Eythórsson and Jónsson2009; Viðarsson 2009; Ingason 2010; Nowenstein 2012, 2014a,b.)

72

(41) {{

Migme.ACC

//

Mérme.DAT

}}

langarwants

aðto

fara.go

‘I want to go.’

However, (41) does not make accusative or dative available in (40B).

And yet, the availability of accusative or dative with a verb like langa ‘want’

does make available a case mismatch of its own: case mismatches based on Dative

Substitution are clearly okay:

(42) A: Migme.ACC

langarwants

aðto

fara.go

‘I want to go.’B: Mér

me.DAT

líka.too

‘Me too.’

(43) A: Hverjumwho.DAT

langarwants

aðto

fara?go

‘Who wants to go?’B: {

{*Ég*I.NOM

//

Migme.ACC

//

Mérme.DAT

}!}

‘Me!’

This is even possible within one sentence, as illustrated in (44a).13 (44b) shows that

such mismatching is not possible with a verb like vilja ‘want’, which, as illustrated

in (40) above, only takes a nominative subject.

(44) a. Hanaher.ACC

langarwants

aðto

fara,go,

ogand

honumhim.DAT

líka.too

‘She wants to go, and he does too.’b. Hún

she.NOM

villwants

fara,go

ogand

{{

hannhe.NOM

//

*honum*him.DAT

}}

líka.too

‘She wants to go, and he does too.’13Note that although (44) is translated using verb phrase ellipsis, that is not what is going on in

the Icelandic examples, as Icelandic doesn’t even have verb phrase ellipsis (Thoms, 2012).

73

It is not enough that a verb can assign two different cases (to the same argu-

ment), however. What appears to be crucial in allowing case mismatches is that in

the case of Dative Substitution, the change in case has no semantic consequences.

It has long been known that some kinds of case alternations do have semantic

consequences. For example, there is a class of verbs which can take either an ac-

cusative or a dative object (H.Á. Sigurðsson, 1989; Barðdal, 1993; Maling, 2002;

Svenonius, 2002). If the dative is chosen, the object is understood to benefit from

the event. Consider the example in (45):

(45) Húnshe.NOM

klóraðiscratched

{{

migme.ACC

//

mérme.DAT

}}

‘She scratched me.’

If accusative is chosen, it means she affected me physically, and probably hurt me

or damaged my skin. If dative is chosen, it means I benefitted from the event, as if

she had scratched me kindly or scratched an itch. With case alternations like this,

a case mismatch in fragment answers is not possible.

(46) A: Húnshe.NOM

klóraðiscratched

mig.me.ACC

‘She scratched me.’B: {

{*Ég*I.NOM

//

Migme.ACC

//

*Mér*me.DAT

}}

líka.too

‘Me too.’

(47) A: Húnshe.NOM

klóraðiscratched

mér.me.DAT

‘She scratched me.’B: {

{*Ég*I.NOM

//

*Mig*me.ACC

//

Mérme.DAT

}}

líka.too

‘Me too.’

Another example comes from cases discussed by Jónsson (2013a), drawing

in part on the references above. Jónsson (2013a) noticed that verbs of contact, like

74

skalla ‘(hit with one’s) head’, can take either accusative or dative objects.14

(48) MessiMessi

skallaðiheaded

{{

boltannball.the.ACC

//

boltanumball.the.DAT

}}

íin

netið.net.the

‘Messi headed the ball into the net.’ (Jónsson, 2013a, 145)

According to Jónsson (2013a, 154), “While both the accusative and the dative vari-

ant assert contact with the object, only the latter variant asserts motion of the ob-

ject.” Thus, a sentence with the dative entails the corresponding sentence with an

accusative, but not vice-versa.

(49) a. JónJohn.NOM

skallaðiheaded

boltannball.the.ACC

án þesswithout

aðto

skallahead

honumit.DAT

neitt.anywhere

‘John headed the ball without heading it anywhere.’b. * Jón

John.NOM

skallaðiheaded


(burt)(away)

án þesswithout

aðto

skallahead

hann.it.ACC

‘John headed the ball away without heading it.’(Jónsson, 2013a, 155)

Similarly to klóra ‘scratch’ above, case mismatches with skalla ‘head’ are not pos-

sible (although the contrast is perhaps sharper with klóra ‘scratch’ than with skalla

‘head’, as pointed out to us by Jóhannes Gísli Jónsson).15

14Note, however, that not all speakers accept dative in sentences like (48). The tests based on it,therefore, can only be judged by speakers who do accept both dative and accusative.

15However, Hlíf Árnadóttir (p.c.) tells us that she finds the following exchange acceptable.

(i) A: JónJohn.NOM

mokaðishoveled

stéttina.sidewalk.the.ACC

‘John shoveled the sidewalk.’B: Já,

Yes,snjónumsnow.the.DAT

líka.too

‘Yes, the snow too.’

Since the alternation between accusative and dative in this case is thought to be determined se-mantically (Svenonius, 2002), on par with the skalla ‘head’ examples above, this suggests thatcase-mismatches might be slightly less restricted than we are indicating here. For now, we leave a

75


skallaðiheaded

fjólubláapurple

boltanum.ball.the.DAT

‘John headed the purple ball.’B: {

{??Gráa??gray

boltannball.the.ACC

//

Gráagray


}}

líka.too

‘The gray ball too.’


skallaðiheaded

fjólubláapurple

boltann.ball.the.ACC


{Gráagray

boltannball.the.ACC

//

??Gráa??gray


}}

líka.too


The contrast between Dative Substitution, which seems to trigger/allow case-

mismatches, and the other cases, which do not, seems to relate to the fact that in

the latter cases, a difference in case assignment correlates with a difference in in-

terpretation, whereas in the former case, it does not (though see footnote 15). This

lends itself to an account which takes the difference in semantic interpretation to

have syntactic correlate, whereas Dative Substitution is a purely morphological

process. Consider why. In the case where a case difference entails a semantic dif-

ference, there must be some marking in the syntax that the semantics is drawing

from. If the case alternation were a purely PF process, there would be no way for

the semantics to be directly sensitive to it. In contrast, Dative Substitution could be

a purely PF process, since the semantics needn’t be sensitive to it.

Icelandic, then, can be seen as a mixed language with respect to whether it

robustly supports or counterexemplifies the CMG—it does both. In the following

section we discuss the theoretical consequences of these facts for extant theories

of ellipsis and ellipsis identity, and the standard conception of the CMG, in light

of the Icelandic counterexamples, and supporting examples just discussed.better understanding for future research.

76

6 Towards an Explanation, and Theoretical Consequences

The facts above receive the most natural explanation if TP ellipsis requires syn-

tactic identity of the ellipsis site with an antecedent, but that the factors deciding

between accusative and dative are not encoded in the syntax. For example, sup-

pose that the experiencer of a verb like langa ‘want’ is introduced in the specifier

of an Appl(icative) head, as proposed in Wood (2015). In many cases, the specifier

of experiencer Appl is assigned dative. However, there is another variety of Appl,

even with an experiencer interpretation, which does not assign dative to its spec-

ifier. H.Á. Sigurðsson (2012) gives the following notation: Appl* assigns dative,

while Appl*+ assigns accusative.16 H.Á. Sigurðsson (2012) argues, however, that

the distinction between Appl* and Appl*+ is made post-syntactically, in the PF

branch of the derivation.

If so, then the rule adding the marked, ‘+’ feature to Appl* would only apply

to a certain, limited set of verbs. Failure to apply this extra, marked rule would lead

to dative rather than accusative.

Another possibility is that there is no syntactic featural difference between

the two varieties of Appl. Rather, in the spirit of McFadden (2004, 2006), there

would be a general post-syntactic rule to the effect that dative case is added to a

DP base-generated in SpecApplP. (See McFadden (2004) for a formalization of

how the case feature assigned to a DP will be realized on all the appropriate heads

internal to that DP.)

(52) DP → DPDAT / [ApplP __ [Appl’ . . . ] ]

Accusative subjects would then involve some way of suppressing the rule in (52)

for particular verbs. One way of suppressing such a rule, which retains the intuition16Wood (to appear) proposes that accusative subjects are in fact structural accusative, assigned

not by Appl itself but by virtue of the presence of a silent external argument. The difference doesnot matter for the present point, though, except that if there is a silent external argument present, itwould have to be present in both the accusative and dative variants, at least for syntactic identity tohold in ellipsis contexts.

77

that (52) is the general case, would be to apply the rule everywhere, but then apply

an impoverishment rule in the context of certain, specific verbs.

(53) DPDAT → DP / __ {langa ‘want’, etc. }

In the absence of a dative feature, accusative will appear automatically in the ac-

count of Wood (to appear). Alternatively, we could specify accusative directly, as

in McFadden (2004). McFadden (2004) argues that dative is really [+OBLIQUE,

+INFERIOR], while accusative is simply [+INFERIOR]. He proposes that Dative

Substitution is a result of the following impoverishment rule:

(54) [+CASE, +OBLIQUE, +INFERIOR] → [+CASE, +INFERIOR] / __ {langa‘want’, etc. }

The general situation would be to add dative (i.e. [+CASE, +OBLIQUE, +INFERIOR])

to a DP in SpecApplP, but in the context of particular verbs—which speakers would

have to learn individually—the [+OBLIQUE] feature would be deleted. This cap-

tures the fact that speakers really do have to learn on a case-by-case basis which

verbs take accusative, but the general system pushes in the direction of dative for

applied arguments.

An advantage to this approach is that it has a clear way of approaching some

other case-mismatches, such as those discussed in detail by Jónsson (2013b). He

discusses examples where DP modifiers carry a different case from the DP they

modify. An example is presented in (55) and (56). Ordinarily, the intensifier ‘self’

must match the DP it modifies in case (and number/gender as well). The examples

in (55) are, in this respect, what we would expect. (55a) would be the standard

variant, with accusative on the subject pronoun, and a matching accusative on the

‘self’ modifier. (55b) would be the expected form in the context of Dative Substi-

tution (which langa ‘want’ permits): dative on the subject pronoun and a matching

dative on the ‘self’ modifer. These examples, are, as we expect, both possible.

78

(55) a. Migme.ACC

sjálfanself.ACC

langarwant

aðto

vitaknow

það.that

‘I myself want to know that.’b. Mér

me.DAT

sjálfumself.DAT

langarwant

aðto

vitaknow

það.that

‘I myself want to know that.’

However, Jónsson (2013b) also claims that case-mismatches are possible, but with

a twist: an accusative subject pronoun is possible with a dative ‘self’ modifier, but

a dative subject pronoun is highly degraded with an accusative ‘self’ modifier.

(56) a. Migme.ACC

sjálfumself.DAT

langarwant

aðto

vitaknow

það.that

‘I myself want to know that.’b. ?? Mér

me.DAT

sjálfanself.ACC

langarwant

aðto

vitaknow

það.that

‘I myself want to know that.’

Jónsson (2013b) presents several other, similar examples, with floating quantifiers

and secondary predicates of various sorts. The approach here would suggest an

asymmetry in how the impoverishment rule that turns underlyingly dative DPs into

accusative ones applies. Essentially, there would be some rule stating that impover-

ishment must apply to the head DP before it can apply to any of the dependents that

agree with it. We must leave a more in-depth development of this idea for future

work, noting only that treating Dative Substition as the absence of impoverishment

seems promising.17

Under all of the above possibilities, however, the narrow syntax makes no

distinction between dative and accusative. As far as the syntax is concerned, the17Another possibility is that the agreeing dependents themselves get their case features under

agreement at PF, in which case the variation could be a matter of timing: either agreement occursbefore impoverishment, generating (56a), or after impoverishment, generating (55a), or else thereis no impoverishment, generating (55b). The marked, degraded option in (56b), however, is theonly option that would require impoverishment directly on the modifier—and that in the absenceof impoverishment on the head.

79

structure is the same whether the DP ended up being dative or accusative. This

structure is enough to license ellipsis. Consider (43), repeated in (57).

(57) A: Hverjumwho.DAT

langarwants

aðto

fara?go

‘Who wants to go?’B: {

{*Ég*I.NOM

//

Migme.ACC

//

Mérme.DAT

}!}

‘Me!’

Speaker A’s utterance would have the structure in (58).

(58) CP

hverjumi

‘who.DAT’TP

tiT

langark‘wants’

. . .vP

vtk

ApplP

tiAppl CP

að fara‘to go’

Speaker B’s utterance would be identical, except that the specifier of Appl would

be a 1st person pronoun. The case of that pronoun would be determined post-

syntactically, on the basis of the surrounding structure—dative because it is in the

specifier of ApplP, possibly followed by accusative (under impoverishment or case-

star augmentation) due to the presence of the specific verb langa ‘want’. But the

syntax of the TP would be the same: not even the features, let alone the structure

of ApplP and its arguments would be different in the narrow syntax.

Now consider the other kind of case-alternation, the one where mismatches

80

are not possible in fragment answers. We repeat examples in (50)–(51) above in

(59)–(60).


skallaðiheaded

fjólubláapurple

boltanum.ball.the.DAT


{??Gráa??gray

boltannball.the.ACC

//

Gráagray


}}

líka.too



skallaðiheaded

fjólubláapurple

boltann.ball.the.ACC


{Gráagray

boltannball.the.ACC

//

??Gráa??gray


}}

líka.too


In this example, the case distinction makes a semantic difference, even if it is a

very subtle one. This would mean that there would have to be some feature in

the syntax that distinguished between them. For example, Schäfer (2008) proposes

that the dative case version involves a special VoiceDAT head. Svenonius (2006),

Jónsson (2013b) and Wood (2015) propose that the [DAT] feature is on a special

kind of little v. E.F. Sigurðsson (2015) proposes that the dative version is actually

structurally distinct from the accusative one, in that the dative argument is gener-

ated in a lower position in the tree.18 For all these proposals, however, there is some

syntactic difference between the accusative and dative structures that the semantics

can be sensitive to. Thus, ruling such cases out on the basis of syntactic identity

promises to be relatively straightforward if we assume such semantic differences

entail a syntactic one, a natural assumption.

Note, in passing, that this does not entail that case is assigned in the syntax.

Rather, post-syntactic case assignment can be sensitive to the presence of, say vDAT

18Jónsson (2013b) also proposes that the dative version is structurally distinct from the ac-cusative, but for him, it is not about the position of the object: rather his v-DAT occurs as an extrahead in addition to the heads present in the accusative version.

81

or VoiceDAT. In fact, Wood (2015) argues that certain case alternations between da-

tive and nominative should be accounted for with an impoverishment rule applied

to vDAT, deleting the [DAT] feature prior to morphological case-assignment. This

account requires that case assignment takes place post-syntactically, even in this

case where a syntactic diacritic on v is necessary.

The claim, then, would be that whatever feature or structure distinguished

accusative from dative in the skalla ‘head’-type examples, that is enough to bleed

ellipsis. The structural distinction—even if it is just a case-feature diacritic on a

head—prevents TP deletion of a structure that differs (e.g. in lacking that feature,

or in the more structurally complex way proposed by E.F. Sigurðsson 2015).

Finally, we point out a theoretical consequence of the Icelandic case facts

for non-silent structure theories (Ginzburg and Sag 2000; Culicover and Jackend-

off 2005; Nykiel and Sag 2012; Barker 2013; Jacobson 2013). Case mismatch

under ellipsis in Icelandic directly challenges “non-silent-structure” approaches,

which reject the notion that there is regular, albeit unpronounced syntactic struc-

ture. Such approaches capture case matching by stipulating that the category of the

remnant and correlate must match, with case features forming part of the category

definition. That is, a nominative marked nominal would differ in category from an

accusative marked one. This assumption leads us to expect case-matching across

the board in Icelandic, contrary to fact.19

On the other hand, the standard account of case matching effects, which

makes reference to silent structure, does a better job of correctly predicting the

distribution of case mismatches we see in Icelandic. Crucial reference is made to

the idiosyncratic properties of elided material. In sluices involving dative substi-

tution predicates or otherwise, it is not the correlate that determines case marking

possibilities for the remnant, but rather the copy of the antecedent predicate in the

elided TP.19Barros (2014) adopts the standard assumption that there is silent structure in ellipsis, but adopts

a semantic theory of identity with an additional case-matching stipulation in the spirit of non-silentstructure approaches. The Icelandic facts would seem to argue against even this view of identity.

82

7 Conclusion

In this paper, we have explored the basic facts of Icelandic clausal ellipsis. For

the most part, Icelandic clausal ellipsis is unexceptional in having the properties

we expect from such constructions as found in many other languages. Independent

properties of the Icelandic case system, however, shed potentially important light

on the analysis of clausal ellipsis.

With respect to apparent cases of island repair under clausal ellipsis, the

Icelandic facts support the view where such repair is not apparent, or illusory, as

has been argued in recent work. To support this conclusion we drew on interpretive

and morphological evidence in controlling for alternative non-island containing

structures, such as short sources and copular clauses. Even when such structures

are independently ruled out, island repair effects persist. It is always possible that

there is some other, non-island-violating alternative at work, but our investigation

so far casts doubt on some of the more prominent proposals.

Icelandic behaves largely as expected with respect to case matching effects

in clausal ellipsis, requiring case on remnants and correlates to match in general.

Case matching is standardly assumed to follow from two assumptions: first, that

there is silent syntactic structure in ellipsis, and second, that this structure is iden-

tical to the structure of the antecedent for the ellipsis.

Together these two assumptions make a simple prediction, borne out in Ice-

landic Dative Substitution configurations, that case mismatches should be possible

when the antecedent predicate (and its copy in the elided structure) allows case to

alternate on one or more of its arguments. The Icelandic facts support the standard

assumptions over approaches that reject silent structure and stipulate case match-

ing between remnants and correlates in clausal ellipsis, with no reference to an

elided predicate (Ginzburg and Sag 2000; Culicover and Jackendoff 2005; Nykiel

and Sag 2012; Jacobson 2013; Barker 2013). Such approaches undergenerate in

overpredicting matching across the board in Icelandic.

83

Importantly, the case matching facts also argue against purely semantic ap-

proaches that allow for structural mismatches between the antecedent and elided

structure (provided they match in interpretation). Such approaches overgenerate,

predicting unacceptable case mismatches. Consider, for instance, the discussion in

section 5 surrounding examples (40) and (41) (repeated below as (61) and (62)).

The verbs vilja and langa both mean ‘want,’ but come with distinct case properties.

A purely semantic approach would predict an antecedent like that in (61A), with

vilja ‘want’ as the main verb, should license deletion in a clause like (62), with

langa ‘want’, erroneously predicting an acceptable accusative or dative remnant.20

(61) a. A: {{

ÉgI.NOM

//

*Mig*me.ACC

//

*Mér*me.DAT

}}

vilwant

fara.go

‘I want to go.’B: {

{ÉgI.NOM

//

*Mig*me.ACC

//

*Mér*me.DAT

}}

líka.too

‘Me too.’

(62) {{

Migme.ACC

//

Mérme.DAT

}}

langarwants

aðto

fara.go

‘I want to go.’

It is worth noting that there is currently no consensus on precisely how the

identity condition on ellipsis should be stated. Here, we have weighed just two

alternatives in broad terms: a purely syntactic condition, and a purely semantic

one. It is well established that either option alone runs into empirical trouble.

Merchant (2001) shows that purely syntactic approaches undergenerate, whereas

Chung (2006) shows that purely semantic approaches overgenerate. Various “hy-

brid” approaches have also been proposed, often adopting an overarching semantic

identity condition alongside one or more syntactic codicils to reign in overgen-

eration (Merchant 2005; Chung 2006, 2013; AnderBois 2011; Barros 2014; Weir20In this case, a semantic account might, however, make use of the more subtle semantic distinc-

tions between langa and vilja discussed by Jónsson (2003, 138). It is well known in the Icelandicliterature, however, that case-marking cannot be predicted on the basis of semantics, so we suspectthat the point here will withstand closer scrutiny.

84

2014). The Icelandic facts we have uncovered are consistent with a purely syntactic

approach, but we appreciate that a hybrid approach may be capable of countenanc-

ing the Icelandic facts as well. For space and time reasons, we leave exploration of

the various hybrid proposals on the market for future research.

References

AnderBois, Scott. 2011. Issues and Alternatives. Doctoral Dissertation, University

of California at Santa Cruz.

Barðdal, Jóhanna. 1993. Accusative and dative case of objects of some transitive

verbs in Icelandic and the semantic distinction between them. In Flyktförsök.

Kalasbok till Christer Platzack på femtioårsdagen 18 november 1993, från

doktorander och dylika, 1–13. Lund: Lunds universitet.

Barðdal, Jóhanna. 2001. Case in Icelandic: A Synchronic, Diachronic, and Com-

parative Approach. Doctoral Dissertation, Lund University.

Barðdal, Jóhanna. 2011. The rise of Dative Substitution in the history of Icelandic:

A diachronic construction grammar account. Lingua 121: 60–79.

Barker, Chris. 2013. Scopability and sluicing. Linguistics and Philosophy 36: 187–

223.

Barros, Matthew. 2012. A non-repair approach to island sensitivity in contrastive

TP ellipsis. In Proceedings of the 48th meeting of the Chicago Linguistic Society,

61–75. Chicago Linguistic Society.

Barros, Matthew. 2014. Sluicing and Identity in Ellipsis. Doctoral Dissertation,

Rutgers, the State University of New Jersey.

Barros, Matthew, Patrick Elliott, and Gary Thoms. 2013. More variation in island

repair: clausal vs. non-clausal islands. In Proceedings of the 49th meeting of the

Chicago Linguistic Society, 331–345. Chicago, IL: Chicago Linguistic Society.

85

Barros, Matthew, Patrick Elliott, and Gary Thoms. 2014. There is no island repair.

Ms. Rutgers/University College London/University of Edinburgh (submitted).

Chung, Sandra. 2006. Sluicing and the lexicon: The point of no return. In BLS 31:

General session and parasession on prosodic variation and change, 73–91.

Chung, Sandra. 2013. Syntactic identity in sluicing: How much and why. Linguistic

Inquiry 44: 1–44.

Chung, Sandra, William A. Ladusaw, and James McCloskey. 1995. Sluicing and

logical form. Natural Language Semantics 3: 239–282.

van Craenenbroeck, Jeroen. 2012. Ellipsis, Identity and Accommodation. Ms. KU

Leuven, HU Brussel.

Culicover, Peter, and Ray Jackendoff. 2005. Simpler Syntax. Oxford: Oxford Uni-

versity Press.

Erteschik-Shir, Nomi. 1977. On the nature of island constraints. Doctoral Disser-

tation, Massachusetts Institute of Technology.

Eythórsson, Thórhallur, and Jóhannes Gísli Jónsson. 2009. Variation in Icelandic

morphosyntax. In Describing and Modeling Variation in Grammar, eds. Andreas

Dufter, Jürg Fleischer, and Guido Seiler, 81–96. Berlin: Mouton de Gruyter.

Fox, Danny, and Howard Lasnik. 2003. Successive cyclic movement and island

repair: the difference between sluicing and VP-ellipsis. Linguistic Inquiry 34:

143–154.

Fukaya, Teruhiko. 2007. Sluicing and Stripping in Japanese and some Implica-

tions. Doctoral Dissertation, University of Southern California.

Gengel, Kirsten. 2007. Focus and Ellipsis: A Generative Analysis of Pseudo-

gapping and other Elliptical Structures. Doctoral Dissertation, University of

Stuttgart.

86

Ginzburg, Jonathan, and Ivan Sag. 2000. Interrogative Investigations: The form,

meaning and use of English interrogatives. Stanford: CSLI Publications.

Griffiths, James, and Anikó Lipták. 2012. Contrast and island sensitivity in clausal

ellipsis. Syntax 17 (3): 661–738.

Hankamer, Jorge. 1971. Deletion in Coordinate Structures. Doctoral Dissertation,

Yale University.

Ince, Atakan. 2012. Sluicing in Turkish. In Sluicing: Cross-linguistic Perspectives,

eds. Jason Merchant and Andrew Simpson 248–269. Oxford: Oxford University

Press.

Ingason, Anton Karl. 2010. Productivity of non-default case. Working Papers in

Scandinavian Syntax 85: 65–117.

Jacobson, Pauline. 2013. The short answer: implications for Direct Composition-

ality. Ms. Brown University.

Jónsson, Jóhannes Gísli. 2003. Not so quirky: On subject case in Icelandic. In New

Perspectives on Case Theory, eds. Ellen Brandner and Heike Zinsmeister, 127–

163. Stanford, CA: CSLI.

Jónsson, Jóhannes Gísli. 2013a. Dative versus accusative and the nature of inherent

case. In Variation in Datives: A Micro-Comparative Perspective, eds. Beatriz

Fernández and Ricardo Etxepare, 144–160. Oxford: Oxford University Press.

Jónsson, Jóhannes Gísli. 2013b. Two types of case variation. Nordic Journal of

Linguistics 36 (1): 5–25.

Jónsson, Jóhannes Gísli, and Thórhallur Eythórsson. 2003. Breytingar á frumlags-

falli í íslensku [Changes in subject case in Icelandic]. Íslenskt mál og almenn

málfræði 25: 7–40.

87

Jónsson, Jóhannes Gísli, and Thórhallur Eythórsson. 2005. Variation in subject

case marking in Insular Scandinavian. Nordic Journal of Linguistics 28 (2): 223–

245.

Lasnik, Howard. 2001. When can you save a structure by destroying it? In Pro-

ceedings of the North East Linguistic Society, vol. 31, eds. Minjoo Kim and Uri

Strauss, 301–320.

Maling, Joan. 2002. Það rignir þágufalli á Íslandi: Sagnir sem stjórna þágufalli

á andlagi sínu [Verbs with dative objects in Icelandic]. Íslenskt mál og almenn

málfræði 24: 31–106.

McFadden, Thomas. 2004. The Position of Morphological Case in the Derivation:

A Study on the Syntax-Morphology Interface. Doctoral Dissertation, University

of Pennsylvania.

McFadden, Thomas. 2006. German inherent datives and argument structure. In

Datives and Other Cases: Between Argument Structure and Event Structure, eds.

Daniel Hole, André Meinunger, and Werner Abraham, 49–77. John Benjamins.

Merchant, Jason. 2001. The Syntax of Silence: Sluicing, Islands, and the Theory of

Ellipsis. Oxford: Oxford University Press.

Merchant, Jason. 2002. Swiping in Germanic. In Studies in Comparative Germanic

Syntax. Proceedings from the 15th Workshop on Comparative Germanic Syntax,

eds. Jan-Wouter Zwart and Werner Abraham, 289–315. Amsterdam: John Ben-

jamins.

Merchant, Jason. 2004. Fragments and ellipsis. Linguistics and Philosophy 27:

661–738.

Merchant, Jason. 2005. Revisiting Syntactic Identity Conditions. Handout from

Workshop on Ellipsis, University of California at Berkeley.

Merchant, Jason. 2008. Variable island repair under ellipsis. In Topics in Ellipsis,

ed. Kyle Johnson, 132–153. Cambridge: Cambridge University Press.

88

Morgan, Jerry. 1973. Sentence Fragments and the Notion “sentence”. In Issues in

Linguistics, eds. Braj Kachru, Robert Lees, Yakov Malkiel, Angelina Pietrangeli,

and Sol Saporta, 719–751. Urbana: University of Illinois Press.

Norris, Mark, Line Mikkelsen, and Jorge Hankamer. 2014. Licensing Trouble. Lin-

guistic Inquiry 45 (4): 617–653.

Nowenstein, Iris Edda. 2012. Mig langar sjálfri til þess. Rannsókn á innri

breytileika í fallmörkun frumlaga. B.A. Thesis, University of Iceland.

Nowenstein, Iris Edda. 2014a. Intra-speaker Variation in Subject Case: Icelandic.

University of Pennsylvania Working Papers in Linguistics 20 (1): 1–10. Available

at: http://repository.upenn.edu/pwpl/vol20/iss1/28/.

Nowenstein, Iris Edda. 2014b. Tilbrigði í frumlagsfalli á máltökuskeiði. Þágufalls-

hneigð og innri breytileiki. M.A. Thesis, University of Iceland.

Nykiel, Joanna, and Ivan Sag. 2012. Remarks on Sluicing. In Proceedings of

HPSG11 Conference.

Ott, Dennis. 2014. An Ellipsis Approach to Contrastive Left-Dislocation. Linguis-

tic Inquiry 45 (2): 269–303.

Ott, Dennis, and Mark de Vries. 2016. Right-dislocation as deletion. Natural Lan-

guage & Linguistic Theory 34: 641–690.

Platzack, Christer. 2008. Cross Linguistic Variation in the Realm of Support Verbs.

Comparative Germanic Syntax Workshop. Available on lingbuzz/000766.

Rögnvaldsson, Eiríkur. 1982. We need (some kind of a) rule of conjunction reduc-

tion. Linguistic Inquiry 13: 557–561.

Rögnvaldsson, Eiríkur. 1990. Null objects in Icelandic. In Modern Icelandic Syn-

tax, eds. Joan Maling and Annie Zaenen, 367–379. San Diego: Academic Press.

Rögnvaldsson, Eiríkur. 1993. Coordination, ATB-extractions, and the identification

of pro. Harvard Working Papers in Linguistics 3: 153–180.

89

Ross, John R. 1967. Constraints on Variables in Syntax. Doctoral Dissertation,

MIT.

Ross, John Robert. 1969. Guess Who? In Papers from the 5th Regional Meeting

of the Chicago Linguistic Society, eds. Robert Binnick, Alice Davison, Georgia

Green, and Jerry Morgan, 252–286. Chicago, IL: Chicago Linguistic Society.

Schäfer, Florian. 2008. The Syntax of (Anti-)Causatives. Philadelphia: John Ben-

jamins.

Sigurðsson, Einar Freyr. 2015. Particle incorporation, resultatives, and ‘re-’ prefix-

ation. Qualifying Paper, University of Pennsylvania.

Sigurðsson, Einar Freyr, and Brynhildur Stefánsdóttir. 2014. ‘By’-phrases

in the Icelandic New Impersonal Passive. University of Pennsylva-

nia Working Papers in Linguistics 20 (1): 311–320. Available at:

http://repository.upenn.edu/pwpl/vol20/iss1/33.

Sigurðsson, Halldór Ármann. 1989. Verbal Syntax and Case in Icelandic. Doctoral

Dissertation, University of Lund.

Sigurðsson, Halldór Ármann. 2012. Case variation: Viruses and star wars. Nordic

Journal of Linguistics 35 (3): 313–342.

Sigurðsson, Halldór Ármann, and Verner Egerland. 2009. Impersonal null-subjects

in Icelandic and elsewhere. Studia Linguistica 63 (1): 158–185.

Svavarsdóttir, Ásta. 1982. Þágufallssýki: Breytingar á fallnotkun í frumlagssæti

ópersónulegra setninga [Dative sickness: changes in the subject case of imper-

sonal sentences]. Íslenskt mál og almenn málfræði 4: 19–62.

Svenonius, Peter. 2002. Icelandic case and the structure of events. Journal of Com-

parative Germanic Linguistics 5 (1–3): 197–225.

Svenonius, Peter. 2006. Case alternations and the Icelandic passive and middle.

In Passives and Impersonals in European Languages, eds. Satu Manninen, Di-

http://repository.upenn.edu/pwpl/vol20/iss1/33

90

ane Nelson, Katrin Hiietam, Elsi Kaiser, and Virve Vihman. Amsterdam: John

Benjamins. http://goo.gl/Ihgh4S.

Thoms, Gary. 2012. Ellipsis licensing and verb movement in Scandinavian.

Manuscript, University of Glasgow.

Thoms, Gary. 2015. Syntactic identity, parallelism and accommodated an-

tecedents. Lingua 166: 172–198.

Thráinsson, Höskuldur. 1975. Gapping in Icelandic: Functional Explanations and

the No-Ambiguity Condition. Chicago Linguistics Society 11: 604–614.

Vicente, Luis. 2015. Morphological case mismatches under sluicing. Snippets 29:

16–17.

Viðarsson, Heimir Freyr. 2009. Tilbrigði í fallmörkun aukafallsfrumlaga:

Þágufallshneigð í forníslensku? [Variation in oblique subject case: Dative Sub-

stitution in Old Icelandic?]. Íslenskt mál og almenn málfræði 31: 15–66.

Weir, Andrew. 2014. Fragments and Clausal Ellipsis. Doctoral Dissertation, Uni-

versity of Massachusetts, Amherst.

Wood, Jim. 2015. Icelandic Morphosyntax and Argument Structure. Dordrecht:

Springer.

Wood, Jim. To appear. The Accusative Subject Generalization. Syntax.

clausal ellipsis and case (mis)matching in icelandic

Documents